[FFmpeg-user] minterpolate performance & alternative

pdr0 pdr0 at shaw.ca
Fri Jan 29 01:20:39 EET 2021

Mark Filipak (ffmpeg) wrote
> Suppose I explain like this: Take any of the various edge-detecting,
> deinterlacing filters and, for 
> each line-pair (y & y+1), align both output lines (y & y+1) to the mean of
> the input's 
> line(y).Y-edge & line(y+1).Y-edge. To do that, only single line-pairs are
> processed (not between 
> line-pairs), and no motion vector interpolation is needed.

There is no decomb or deinterlacing filter that does this. 

How you would do is deinterlace and optical flow. These are separate

Motion vectors are required, otherwise - how else would you determine where
an object moves ? How else would you get the correct "mean" and position?  A
sobel operator looks at the current frame (or field). It's a spatial
operation. There is no relationship between the previous or next. Or if you
want, even scanlines are independent from odd scanlines.  The relationship
is the  motion vector.

Y and Y+1 are even and odd scanlines.  They are combed in the current frame
because they come from different points in time . T=0, T=1.

In this case, your "spatial mean"  is also  a "temporal mean" . eg. As an
object or pixel moves from position x=1 at T=0 to x=5 at T=1 in the next
field. The spatial mean is x=3 at T=0.5.  Instead of time T=0, T=1, you want
T=0.5, assuming linear interpolation. This is optical flow. The inbetween
point in time, and it's resulting data.

Sent from: http://ffmpeg-users.933282.n4.nabble.com/

More information about the ffmpeg-user mailing list